Eddy Current Array for Defect Detection in Finely Grooved Structure Using MSTSA Network.

Area of Science:

• Non-destructive testing
• Signal processing
• Machine learning for engineering

Background:

• Eddy Current Array (ECA) technology is crucial for defect detection in industrial components.
• Surface texture, lift-off, and mechanical dither pose significant challenges to ECA accuracy.
• Existing methods struggle with the complex interference patterns in finely grooved structures.

Purpose of the Study:

• To develop an advanced ECA-based defect detection method for finely grooved spinning cylinders.
• To address the limitations of traditional methods in handling surface interference and signal noise.
• To introduce a novel deep learning framework for improved defect detection accuracy and efficiency.

Main Methods:

• Proposed a Multi-scale SpatioTemporal Self-Attention Network (MSTSA-Net) for ECA defect detection.
• Incorporated Temporal Attention (TA) and Spatial Attention (SA) blocks to capture spatiotemporal defect features.
• Utilized depth-wise and point-wise convolutions for self-attention weight computation and fused multi-scale features for defect localization.

Main Results:

• MSTSA-Net demonstrated superior performance compared to traditional image processing and state-of-the-art models (YOLOv3-SPP, Faster R-CNN).
• The proposed method achieved higher Recall and F1 scores in defect detection.
• MSTSA-Net requires fewer parameters and lower Floating Point Operations (FLOPs).

Conclusions:

• The developed MSTSA-Net effectively detects defects in challenging finely grooved structures using ECA.
• The spatiotemporal self-attention mechanism enhances the ability to identify defects of various sizes amidst interference.
• MSTSA-Net offers a more efficient and accurate solution for non-destructive testing of grooved cylinders.